In a follow‑up to his MCP integration guide, the author demonstrates how to turn an eight‑year weekly visualization habit into a reusable AI workflow. He defines a “skill” as a packaged set of instructions plus optional assets, anchored by a SKILL.md file. Hosting skills on the public repo skills.sh lets LLMs pull lightweight metadata before loading full logic.

The author built two skills for his weekly pipeline: a storytelling‑viz skill that inspects a dataset, surfaces insights, picks a chart type and emits an interactive graphic, and a viz‑publish skill that pushes the result to his personal site. Triggering the first skill in Codex Desktop reduced a one‑hour manual routine to under ten minutes, delivering an insight‑driven headline, chart, caveats and source link.

Iterating on the initial version revealed gaps—suboptimal chart choices, inconsistent styling, and missing takeaways. By feeding the model eight years of personal design guidelines, scraping public visualization resources, and testing across 15+ datasets, the author refined the skill to standardize fonts, ensure mobile‑friendly layouts, and always cite data sources. The resulting modular approach shows how MCP and custom skills can streamline recurring data‑science tasks.